try to run train.py meet error

[drsleep]

try to run train.py meet error

[DrSleep]

What kind of data do you use?

In my experience, this error is usually seen when your target masks have labels larger than the channel dimension of your prediction present.
As a sanity check, you can print soft_output.size(1) and torch.unique(target_var) before computing the loss

[csufangyu]

What kind of data do you use?

In my experience, this error is usually seen when your target masks have labels larger than the channel dimension of your prediction present.
As a sanity check, you can print soft_output.size(1) and torch.unique(target_var) before computing the loss

thanks very much! I use NYUv2,but I find 56 classes!
40 56 tensor([ 0, 2, 3, 5, 6, 7, 8, 9, 11, 13, 14, 15, 17, 19, 20, 21, 24, 34, 35, 36, 39, 42, 45, 49, 55, 59, 64, 70, 73, 74, 75, 80, 82, 83, 93, 111, 113, 119, 133, 138, 141, 144, 148, 156, 157, 158, 160, 161, 175, 177, 181, 206, 234, 239, 243, 255], device='cuda:0')

[DrSleep]

everything above 39 should not be there. check out the NYUDv2 instruction from their site - you may need to run a matlab script to convert the labels

I will try to upload the dataset with the correct labels this week

[csufangyu]

everything above 39 should not be there. check out the NYUDv2 instruction from their site - you may need to run a matlab script to convert the labels

I will try to upload the dataset with the correct labels this week

I have solved it !thanks very much!

[DrSleep]

great, no worries!